This invention relates generally to one or more computer networks that include computers, such as personal computers (PC's) or network computers such as servers, which have microprocessors linked by broadband transmission means and have hardware, software, firmware, and other means such that at least two parallel processing operations occur that involve at least two sets of computers in the network or in interconnected networks. This invention constitutes a form of metacomputing.
More particularly, this invention relates to one or more large networks, like the Internet, which comprise smaller networks and large numbers of interconnected computers, wherein multiple separate parallel or massively parallel processing operations involving multiple different sets of computers occur simultaneously. Even more particularly, this invention relates to one or more such networks wherein multiple parallel or massively parallel microprocessing processing operations occur separately or in an interrelated fashion, and wherein ongoing network processing linkages are established between virtually any microprocessors of separate computers connected to the network.
Still more particularly, this invention relates generally to a network structure or architecture that enables the shared use of network microprocessors for parallel processing, including massive parallel processing, and other shared processing such as multitasking, wherein personal computer owners provide microprocessor processing power to a network, such as for parallel or massively parallel processing or multitasking, in exchange for network linkage to other personal computers and other computers supplied by network providers such as Internet Service Providers (ISP's), including linkage to other microprocessors for parallel or other processing such as multitasking. The financial basis of the shared use between owners and providers may be whatever terms to which the parties agree, subject to governing laws, regulations, or rules, including payment from either party to the other based on periodic measurement of net use or provision of processing power like a deregulated electrical power grid or involving no payment. The network system may provide an essentially equivalent usage of computing resources by both users and providers since any network computer operated by either entity is potentially both a user and provider of computing resources alternately or simultaneously, assuming multitasking is operative. A user may have an override option exercised on the basis of, for example, a user profile, a user's credit line, or relatively instant payment.
This invention also relates to a network system architecture including hardware and software that provides use of the Internet or other network, without cost, to users of personal computers or other computers, while also providing users with computer processing performance that at least doubles every 18 months through metacomputing means. This metacomputing performance increase provided by the new Grid (or Metalnternet) is in addition to other performance increases, such as those already anticipated by Moore's Law.
The computer industry has been governed over the last 30 years by Moore's Law, which holds that the circuitry of computer chips shrinks substantially each year, yielding a new generation of chips every 18 months with twice as many transistors, such that microprocessor computing power effectively doubles every year-and-a-half.
The long-term trend in computer chip miniaturization is projected to continue unabated over the next few decades. For example, slightly more than a decade ago a 16 kilobit DRAM (dynamic random access memory) memory chip (storing 16,000 data bits) was typical; the standard in 1996 was the 16 megabit chip (16,000,000 data bits), which was introduced in 1993; industry projections are for 16 gigabit memory chips (16,000,000,000 data bits) to be introduced in 2008 and 64 gigabit chips in 2011; and 16 terabit chips (16,000,000,000,000 data bits) may be conceivable by the mid-to-late 2020's, by which time such microchips may have become nanochips in terms of their circuit dimensions. This is a thousand-fold increase regularly every fifteen years. Hard drive speed and capacity are also growing at a spectacular rate, even higher in recent years than that of semiconductor microchips.
Similarly, regular and enormous improvements may continue in microprocessor computing speeds, whether measured in simple clock speed or MIPS (millions of instructions per second) or numbers of transistors per chip. For example, performance has improved by four or five times every three years since Intel launched its X86 family of microprocessors used in the currently dominant “Wintel” standard personal computers. The initial Intel Pentium Pro microprocessor was introduced in 1995 and is a thousand times faster than the first IBM standard PC microprocessor, the Intel 8088, which was introduced in 1979. By 1996 the fastest of microprocessors, such as Digital Equipment Corporation's Alpha chip, and even the microprocessor of the Nintendo 64 video game system, were faster than the processor in the original Cray Y-MP supercomputer.
Microprocessors, software, firmware, and other components are also evolving from 8-bit and 16-bit systems into the 32-bit systems that are becoming the standard today, with some 64-bit systems like the DEC Alpha already introduced and more coming, such as Intel's Itanium microprocessor in 2001, with future increases to 128-bit systems likely.
A second major development trend in the past decade or so has been the rise of parallel processing, a computer architecture utilizing more than one CPU microprocessor linked together into a single computer with new operating systems having modifications that allow such an approach. Thousands of relatively simple microprocessors may be used together for massively parallel processing. The field of supercomputing has been overtaken by this approach, which includes designs utilizing many identical standard personal computer microprocessors.
Hardware, firmware, software, and other components specific to parallel processing are in a relatively early stage of development compared to that for single processor computing. Therefore, much further design and development are expected in the future to better maximize the computing capacity made possible by parallel processing. Continued improvement is anticipated in system hardware, software, and architectures for parallel processing so that reliance on the need for multiple microprocessors to share a common central memory is reduced, thereby allowing more independent operation of those general purpose microprocessors, each with their own discrete memory, like current personal computers, workstations, and most other computer systems architecture. For unconstrained operation, each individual microprocessor should have rapid access to sufficient memory.
Several models of personal computers having more than one general purpose microprocessor are now available. In the future, personal computers, broadly defined to include versions not currently in use, will likely also employ parallel computing utilizing multiple microprocessors or massively parallel computing with very large numbers of microprocessors. Future designs, such as Intel's Itanium chip, are expected to have a significant number of parallel processors on a single microprocessor chip.
A form of parallel processing called superscalar processing is also being employed within microprocessor design. The current generation of microprocessors, such as the Intel Pentium, have more than one data path within the microprocessor in which data is processed, with two to three paths being typical now and as many as eight in 1998 in IBM's new Power 3 microprocessor chip.
A third major development trend is the increasing size of bandwidth, which is a measure of communications power or transmission speed, in terms of units of data per second, between computers connected by a network. Previously, the local area networks and telephone lines typically linking computers including personal computers have operated at speeds much lower than the processing speeds of a personal computer. For example, a typical 1997 Intel Pentium operates at 100 MIPS, whereas the most common current Ethernet connecting PC's is roughly 10 times slower at 10 megabits per second (Mbps), although some Ethernet connections are now 100 Mbps and telephone lines are very much slower, the highest typical speed in 1998 being the approximately 56 kilobits reachedduring downloads.
The situation is expected to change dramatically. Bandwidth or transmission speed is anticipated to expand from 5 to 100 times as fast as the rise of microprocessor speeds, due to the use of coaxial cable, wireless, and especially fiber optic cable and optical wireless, instead of old telephone twisted pair lines, and due to the use of wideband communication such as dense wave division multiplexing (DWDM) and wideband code division multiple access (CDMA), as well as ultrawideband wireless. In DWDM systems, multiple channels are transmitted over a single fiber because they are sent at different wavelengths. Telecommunication providers are now making available single fiber connections supporting a bandwidth of 40 gigabits per single fiber, and, alternatively, as many as 160 wavelength channels (lambdas) per single fiber. In CDMA systems, users are multiplexed across the same spectrum, with each user being assigned a different instance of a noise-like carrier wave.
Technical improvements are expected in the near term which will make it possible to carry over 2 gigahertz (billions of cycles per second) on each of 700 wavelength channels (lambdas), adding up to more than 1,400 gigahertz on a single fiber thread. Experts have estimated that the bandwidth of optical fiber has been utilized one million times less fully than the bandwidth of coaxial or twisted pair copper lines. Within a decade, 10,000 wavelength streams per fiber are expected; 20 to 80 wavelengths on a single fiber is already commercially available. The use of thin mirrored hollow wires or tubes called omniguides may also provide very substantial additional increases.
Other network connection developments, such as asynchronous transfer mode (ATM) and digital signal processors, whose price/performance ratio has improved tenfold every two years, are also supporting the rapid increase in bandwidth. The increase in bandwidth reduces the need for switching, and switching speed will be greatly enhanced when practical optical switches are introduced in the near future, potentially reducing costs substantially.
The result of this huge bandwidth increase is extraordinary: already it is technically possible to connect virtually any computer to a network with a bandwidth that equals or exceeds the computer's own internal system bus speed, even as that bus speed itself is increasing significantly. The principal constraint is the infrastructure, consisting mostly of connecting the “last mile” to personal computers with optical fiber or other broad bandwidth connections, which still need to be built. The system bus of a computer is its internal network connecting many or most of its internal components such as microprocessor, random access memory (RAM), hard drive, modem, floppy drive, and CD-ROM; for recent personal computers, the system bus has been only about 40 megabits per second, but is up to 133 megabits per second on Intel's Pentium PCI bus in 1995. IBM's 1998 Power3 microprocessor chip has a system bus of 1.6 gigabits per second and there is now up to a gigabit per second on Intel's Pentium PCI bus.
Despite these tremendous improvements anticipated in the future, a typical PC is already so fast that its microprocessor is essentially idle during most of the time the PC is in actual use, and the operating time itself is but a small fraction of those days the PC is even in use at all. Nearly all PC's are essentially idle during roughly all of their useful life. A microprocessor of a PC may be in an idle state 99.9% of the time, disregarding unnecessary microprocessor busywork such as executing screen saver programs, which have been made essentially obsolete by power-saving CRT monitor technology, which is now standard in the PC industry.
Because the reliability of PC's is so exceptionally high now, with the mean time to failure of all components typically several hundred thousand hours or more, the huge idle time of PC's represents a total loss; given the high capital and operating costs of PC's, the economic loss is very high. PC idle time does not in effect store a PC, saving it for future use, since the principle limiting factor to continued use of today's PC's is obsolescence, not equipment failure resulting from use.
Moreover, there is continuing concern that Moore's Law, which holds that the constant miniaturization of circuits results in a doubling of computing power every 18 months, cannot continue to hold true much longer. Indeed, Moore's Law may now be nearing its limits for silicon-based devices, perhaps by as early as 2010. No new technologies have yet emerged that seem to have the potential for development to a practical level by then, although many recent advances have the potential to maintain Moore's Law.